Structural Health Monitoring of Wind Turbine Blades by Ice Mass Accumulation Prediction Using Convolutional Neural Networks

The growth of wind energy in regions with extreme climates and low temperatures has underscored the need to monitor and quantify ice formations on wind turbine blades. This study aims to tackle this challenge by developing a deep learning model based on convolutional neural networks (CNNs) capable of predicting ice accumulation on blades using experimental data that replicate real ice formation conditions. The mass distribution applied to the blades is derived from real environmental conditions and ice formation equations on aerodynamic profiles and is tested on a 2.4-m blade of a 5-kW wind turbine under different operating conditions. Acceleration time series recorded on the blade, rotation speed, and system temperature are used as input information for the CNN-based model. The model is trained through exhaustive hyperparameter tuning, and the final parameters are determined via cross-validation to ensure its robustness and generalization capability. The results show that the model predicts the accumulated mass on the blade with high accuracy, achieving a fit with R² = 0.996 and R² = 0.994, a mean absolute percentage error (MAPE) of 0.497 and 0.488, and root mean squared error (RMSE) values of 5.64 and 5.79 for the training and testing data, respectively. The proposed approach is consistent with variations in temperature and rotational speed of the system, making it potentially applicable to instrumented de-icing systems by integrating monitoring data with operational and environmental information.